Digital Scotland: Fig. III.6

Local effects of SIMD factors on the odds of being online.

The effects of each factor vary significantly from one Local Authority area to another.

We use a subset of the training dataset, because there is insufficient data to produce this model for Eilean Siar. The model is produced using the following code:

st <- subset(training, CA !="Eilean Siar")
st$CA <- factor(st$CA)

CA2glm <-
    vglm(formula =
               cbind(pmax(0, Household_Count-connections),
                     pmin(Household_Count, connections))
               ~
               slowlines + isolation + CA
               + CA:(income + employment + housing
                   + health + education + crime),
               family = betabinomial(zero=NULL), 
               data = st, trace = TRUE)

Again, we use the parametrisation of the betabinomial distribution in terms of its mean and dispersion, as this gives coefficients we can interpret directly as an odds ratios.

In order to compare like with like, the offsets included in this diagram are produced using a different model, in which the effects of each factor are the same across all of Scotland:

CAglm <- vglm(formula =
         cbind(offline, connected)
         ~ CA 
         + income + employment + housing + health +
         + education + crime + slowlines + isolation,
         family = betabinomial(zero=NULL), 
         data = st)